PaperlayerView exact revision on arXiv

Source-linked AI summary

Overview of Millimeter Wave Communications for Fifth-Generation (5G) Wireless Networks-with a focus on Propagation Models

Theodore S. Rappaport, Yunchou Xing, George R. MacCartney,, Andreas F. Molisch, Evangelos Mellios, Jianhua Zhang

arXiv:1708.02557v1cs.IT

TL;DR

The paper addresses the need for propagation knowledge as 5G systems move toward mmWave frequencies and their associated channel, penetration, and path-loss challenges. It surveys 5G concepts and compares propagation parameters and channel models from international groups and standards bodies across 0.5–100 GHz. The overview compiles early models and measurements while identifying the need for continued validation of evolving mmWave channel models.

  • Problem

    5G mmWave systems require accurate propagation models because greater diffraction and penetration losses make LOS conditions, path loss, and material effects important for system design.

  • Method

    The paper reviews 5G mmWave system concepts and compiles and compares LOS-probability, large-scale path-loss, and building-penetration models from international groups and standards bodies.

  • Results

    The survey shows that early mmWave models differ across standards and research groups, while measurements and ray-tracing studies provide scenario-specific propagation results, including substantial building penetration losses.

  • Takeaways & Limitations

    Propagation knowledge on path loss and shadowing is presented as a prerequisite for advancing mmWave wireless systems and educating engineers working on the developing 5G industry.

Abstract

from arXiv · show

This paper provides an overview of the features of fifth generation (5G) wireless communication systems now being developed for use in the millimeter wave (mmWave) frequency bands. Early results and key concepts of 5G networks are presented, and the channel modeling efforts of many international groups for both licensed and unlicensed applications are described here. Propagation parameters and channel models for understanding mmWave propagation, such as line-of-sight (LOS) probabilities, large-scale path loss, and building penetration loss, as modeled by various standardization bodies, are compared over the 0.5-100 GHz range.

I. INTRODUCTION

Growing wireless traffic is motivating 5G systems that use mmWave spectrum for multi-Gigabit-per-second rates, denser architectures, and new network technologies. The paper introduces these concepts and frames propagation, backhaul, cell deployment, and air-interface design as central development areas.

  • 5G Motivation: Over 50% annual growth in wireless data traffic is driving 5G development, including mmWave systems targeting multi-Gigabit-per-second rates.A 1 GHz channel at 28 or 73 GHz could offer several Gbps to a mobile device with modest phased-array antennas.
  • Propagation Challenges: MmWave channels will exceed 4G LTE’s 20 MHz bandwidth by more than ten times, while greater attenuation increases the importance of LOS propagation, reflection, and scattering.Shorter wavelengths at mmWave frequencies lead to greater diffraction and material-penetration losses.
  • Paper Scope: The paper surveys 5G concepts, propagation challenges, antenna technologies, and channel models developed by international groups and standards bodies.Its channel-model comparison covers models for emerging mmWave wireless communication networks.
  • Backhaul and Fronthaul: MmWave wireless backhaul and fronthaul are proposed to provide fiber-like rates and bandwidth without deploying wired backhaul or long-range digital radio-over-fiber.Fronthaul connects remote radio heads with centralized baseband units, while backhaul connects baseband units to the core network.
  • Small Cells: Shrinking cell size increases area spectral efficiency by reducing users per cell and enabling greater spectrum reuse, although interference and synchronization become challenges.Future 5G deployments also envision nomadic base stations and direct device-to-device connections for greater capacity per user.
  • Multi-Tier Architecture: A multi-tier architecture combines broad 4G coverage with closely spaced 5G base stations, supporting differentiated access priorities and connection types.The architecture includes macrocells, small cells, and device-to-device connections.

E. 5G Unlicensed WiFi

Unlicensed mmWave WiFi and vehicular applications demonstrate multi-gigabit potential, while propagation challenges—including atmospheric absorption, penetration loss, rapid spatial variation, and uncertain stationarity—shape channel design and modeling.

  • 5G Unlicensed WiFi: IEEE 802.11ad reaches up to 7 Gbps in the 60 GHz band, while WirelessHD products reach 4 Gbps with theoretical rates up to 25 Gbps.Both standards support directional antennas, beamforming, and related enhancements for high-throughput unlicensed mmWave communications.
  • Vehicular Networks: 24 GHz and 77 GHz bands are attractive for vehicular links because connected vehicles require Gbps data rates unavailable in 10 MHz channels at 5.9 GHz.
  • Propagation Challenges: Atmospheric absorption varies strongly with frequency: 183, 325, and especially 380 GHz suffer much greater attenuation, while many mmWave bands add only 1–2 dB/km over free-space loss.Rain can also cause substantial attenuation above 10 GHz; at 73 GHz, attenuation reaches 10 dB/km for a 50 mm/hr rain rate.
  • Propagation Challenges: With constant physical antenna size and clear weather, free-space path loss decreases quadratically as frequency increases, but higher antenna gains require adaptive beam steering.The comparison changes when antenna gain is assumed constant over frequency.
  • Propagation Challenges: Building penetration losses at mmWave can be severe, ranging from 3.6 dB for indoor clear glass to 40.1 dB for outdoor tinted glass at 28 GHz.
  • Propagation Challenges: At 60 GHz, rough-surface scattering can produce fade depths up to 20 dB over movements of only a few centimeters, requiring rapid channel adaptation.These variations affect channel-state feedback, link adaptation, beam tracking, and retransmission-related protocol design.
  • Propagation Challenges: Measurements indicate sharp spatial decorrelation over movements of a few tens of wavelengths, while the validity of existing stationarity statistics for mmWave remains uncertain.The necessity and proper form of spatial consistency have not yet been fully understood.

1) UMi LOS Probability:

UMi LOS probability models differ mainly in parameter values and model form across standardization groups. The 5GCM NYU squared model offers a more realistic decay for urban clutter, while indoor models cover several office layouts.

  • UMi scenarios cover dense open areas and street canyons with BS heights below rooftops, UE heights near 1.5 m, and ISDs up to 200 m.
  • 3GPP TR 38.901: 3GPP TR 38.901 uses the d1/d2 model with a 10 m antenna height and UMi parameters d1 = 18 m and d2 = 36 m.
  • 5GCM: 5GCM provides a d1/d2 model and the NYU squared model; the latter has more realistic and rapid decay over distance for urban clutter.The 5GCM d1/d2 model has slightly smaller MSE, whereas the NYU squared model better reflects urban-clutter decay.
  • Other groups: METIS uses the same LOS model form and parameter values as 3GPP TR 38.901, while mmMAGIC uses the 5GCM d1/d2 model.
  • InH: Indoor-office LOS models distinguish mixed and open offices, while 5GCM also models open-plan, closed-plan, and hybrid-plan layouts using ray-tracing simulations.

4) RMa LOS Probability:

The RMa section presents LOS modeling alongside the path-loss model families used across mmWave scenarios. It also identifies scope and validation limits, including rural models derived from urban sub-6 GHz data and omnidirectional models that require directional-channel information for antenna analysis.

  • RMa LOS probability: RMa LOS probabilities were not specified in METIS or 5GCM; 3GPP TR 38.901 adopted its model from ITU-R M.2135 and WINNER.
  • RMa LOS probability: The RMa model defines P_LOS for a T-R pair as a function of 2D separation distance, but caution is advised because its basis is urban data below 6 GHz.
  • Path-loss models: Three basic large-scale path-loss families are identified: CI, frequency- or height-weighted CI variants, and FI/ABG.
  • Scope limitation: Omnidirectional path-loss models cannot be directly used for directional antenna analysis without antenna patterns and properly modeled spatial and temporal multipath statistics.
  • CI model: The CI model anchors distance-dependent loss to 1 m free-space path loss, with the path-loss exponent fitted to measured data.
  • CIF and CIH models: The CIF model adds frequency dependence to the path-loss exponent, while CIH adapts that form to BS-height dependence in RMa scenarios.

P LCIF

The dual-slope path-loss formulation models different distance regimes around a breakpoint. It uses separate slopes before and after that breakpoint while retaining frequency and offset parameters.

  • Dual-slope CIF: The dual-slope CIF model changes its distance-dependent formulation at the breakpoint distance d_BP.
  • Dual-slope ABG: The dual-slope ABG model uses one distance slope before d_BP and another after d_BP, with continuity enforced at the breakpoint.
  • Parameterization: Both dual-slope models require five parameters to predict distance-dependent average path loss.

1) UMi Large-Scale Path Loss:

UMi path-loss models span CI, ABG, breakpoint, and modified sub-6 GHz formulations across LOS and NLOS conditions. The overview compares these models and highlights uncertainty about breakpoint behavior at mmWave frequencies.

  • 5GCM: 5GCM uses the CI model for UMi LOS because its exponent aligns with the ABG exponent and remains close to the free-space value of 2.Both CI and ABG models are also adopted for UMi NLOS.
  • 3GPP TR 38.901: 3GPP TR 38.901 models UMi LOS with a breakpoint formulation based on 3D distance, carrier frequency, and antenna heights.Before the breakpoint, the model is essentially a CI model with n = 2.1.
  • Breakpoint behavior: At mmWave frequencies, breakpoint existence is controversial because measurements have not reported it, although some ray-tracing simulations predict one.For typical UMi cells of 500 m or less, the breakpoint may exceed the cell range, making the simpler CI model similar in prediction.
  • Model comparison: METIS uses a modified ITU-R UMi model claimed valid from 0.8 to 60 GHz, while mmMAGIC adopts an ABG model with different parameters.Comparisons among the UMi large-scale path-loss models are presented in Fig. 5.

2) UMa Large-Scale Path Loss:

UMa path-loss modeling combines CI, CIF, and ABG formulations across several standards and projects. The overview also identifies limited measurement validation in some 3GPP cases.

  • 3GPP TR 38.901: 3GPP TR 38.901 models UMa omnidirectional path loss from 0.5-100 GHz but lacks measurement validation in some cases.Its UMa LOS model reverts to a model defined below 6 GHz and omits the InH shopping mall scenario used in TR 38.900.
  • Overall comparison: The UMa comparison therefore includes models inherited from sub-6 GHz standards alongside CI, CIF, and ABG alternatives.These formulations differ in their parameterizations and stated validation support.
  • 5GCM: 5GCM provides CI, CIF, and ABG models for UMa, with lower CI/CIF path-loss exponents than in UMi.The paper relates this difference to the larger BS height and fewer obstructions in UMa.
  • METIS: METIS adopts the sub-6 GHz 3GPP TR 36.873 3D UMa model published for LTE.The adopted model is listed among the UMa path-loss models compared in Table V.

3) InH Large-Scale Path Loss:

Indoor hotspot models use CI, CIF, ABG, dual-slope, and geometry-based formulations for LOS and NLOS propagation. Their assumptions and parameter complexity vary across standards and projects.

  • 5GCM: 5GCM proposes CI, CIF, ABG, and dual-slope models for InH propagation, including five-parameter dual-slope models and a two-parameter single-slope CIF model.The dual-slope ABG and CIF models are considered for NLOS performance evaluation.
  • Model complexity: The 5GCM data are cited as showing that the additional complexity of some indoor models is warranted compared with the simple CIF model.The passage does not specify the underlying metric or comparison value.
  • 3GPP TR 38.901: 3GPP TR 38.901 claims its InH-office LOS model is valid up to 100 m and uses the same form as the UMi CI model.Its slightly lower exponent is attributed to indoor reflections, scattering, and waveguiding effects.
  • 3GPP TR 38.901: The 3GPP InH-office NLOS model uses an ABG form without a height correction term and applies a patch to keep it below-bounded by LOS path loss.Its stated expression is PLInH-NLOS [dB] = 17.30 + 38.3 log10(d3D) + 24.9 log10(fc).
  • METIS: METIS adopts the WINNER II path-loss model as a geometry-based stochastic model for short-range 60 GHz indoor links.The model uses curve-fit parameters without Friis’ equation.
  • IEEE 802.11ad: IEEE 802.11ad models LOS STA-STA path loss with the CI form of Friis’ free-space equation and reports NLOS FI/AB and ABG variants for STA links.The STA-STA NLOS shadow-fading standard deviation is 3.3 dB, while STA-AP NLOS uses 3.0 dB.

4) RMa Large-Scale Path Loss:

RMa path-loss models largely derive from sub-6 GHz formulations, while alternative CIH models use broader propagation evidence. The overview emphasizes applicability and validation limits for rural mmWave modeling.

  • ITU-R: The ITU-R RMa LOS model uses breakpoint and correction-factor terms, with street width, building height, BS height, and UE height among its parameters.Its stated maximum 2D LOS separation distance is 10 km, while NLOS reaches 5 km.
  • Applicability boundary: At 9.1 GHz or above, the ITU-R LOS model becomes a single-slope model because its breakpoint exceeds the 10 km applicability limit.The paper describes this as mathematically inconsistent for mmWave frequencies above 9.1 GHz.
  • Validation limits: The ITU-R RMa NLOS model is specified only up to 6 GHz and has not been validated in the literature for mmWave frequencies.The paper also states that its measurements came from downtown Tokyo rather than rural scenarios.
  • NYU RMa model: NYU proposed empirically based CIH RMa LOS and NLOS models using extensive simulations and 73 GHz field data.The stated LOS and NLOS shadow-fading standard deviations are 1.7 dB and 6.7 dB, respectively.

C. O2I Penetration Loss

3GPP TR 38.901 models outdoor-to-indoor (O2I) path loss as outdoor loss plus building-wall penetration and indoor depth-dependent loss. Its penetration treatment accounts for wall materials, incidence angle, scenario-specific models, and car-window conditions.

  • 3GPP TR 38.901: O2I path loss combines basic outdoor loss, external-wall penetration loss, indoor depth-dependent loss, and penetration-loss variability.The model explicitly separates PLb, PLtw, PLin, and σP.
  • 3GPP TR 38.901: External-wall loss includes an added non-perpendicular-incidence term and material losses weighted by each material’s proportion.Material loss is modeled as L_materiali = a_materiali + b_materiali · f_c, with material proportions summing to one.
  • 3GPP TR 38.901: Penetration-loss values for different materials and O2I models are compiled in Table IX.The table is specifically identified as covering material penetration loss and O2I penetration-loss models.
  • 3GPP TR 38.901: Both low-loss and high-loss rough models apply to UMa and UMi-street-canyon scenarios, whereas RMa uses only the low-loss model.The scenario restriction is stated for the rough building-penetration models.
  • 3GPP TR 38.901: For most cases, car penetration uses μ = 9 dB and σP = 5 dB, with μ = 20 dB optionally specified for metalized windows from 0.6 to 60 GHz.The car O2I model is included in path loss and distinguishes metalized car windows.

2) 5GCM:

5GCM adopts a building-penetration model derived from 3GPP TR 36.873 and legacy measurements below 6 GHz. It also describes frequency-dependent alternatives and selects standard deviation from measurement data.

  • 5GCM: 5GCM adopts the 3GPP TR 36.873 building-penetration model, which is based on legacy measurements below 6 GHz.The model’s measurement basis limits the stated empirical foundation to frequencies below 6 GHz.
  • 5GCM: 5GCM-related work also proposes several frequency-dependent building-penetration models and a composite external-wall approach.These alternatives are attributed to prior studies and a detailed external-wall treatment, respectively.
  • 5GCM: For the difference between 5GCM and 3GPP TR 38.901, the penetration-loss standard deviation is tentatively selected from measurement data.The passage identifies standard-deviation selection as the stated distinction.
  • 5GCM: The 5GCM low-loss building model uses A = 5 and B = 0.03, while its high-loss model uses A = 10 and B = 5.These parameters are given for low-loss and high-loss buildings, respectively.

3) mmMAGIC:

mmMAGIC expresses O2I penetration loss and shadow fading with frequency-dependent logarithmic models, while the overview contrasts these statistical models with spatially consistent channel-modeling approaches. The paper concludes that propagation models are prerequisites for advancing mmWave systems.

  • 3) mmMAGIC: mmMAGIC models O2I penetration loss as BO2I + CO2I · log10(f_c), approximately 8.5 + 11.2 · log10(f_c).The coefficients can be incorporated into existing mmMAGIC path-loss coefficients.
  • 3) mmMAGIC: For UMi-O2I, frequency-dependent shadow fading is modeled as approximately 5.7 + 2.3 · log10(f_c), with values between 8 and 10 dB presented.The passage presents both the logarithmic expression and the stated shadow-fading range.
  • Spatial consistency: Drop-based models randomly place UEs and assign channel parameters, but they do not provide spatial consistency for nearby transmitter–receiver configurations.They remain useful for statistical or Monte Carlo analysis and may involve short movements up to 40 wavelengths.
  • Spatial consistency: 5GCM and 3GPP generate spatially filtered LOS and shadowing maps that are reused across UE trajectories, while METIS and MiWEBA provide spatial consistency inherently.5GCM and 3GPP also add procedures for consistent delays and angles, whereas those details are outside the overview’s scope.
  • Modeling scope: Standardization choices may reflect legacy-software compatibility and convergence goals in addition to physical laws or fit to measured channel characteristics.The paper describes 5G channel modeling as ongoing and notes early capacity differences among models.
  • Conclusion: The overview compiles mmWave propagation models from independent groups using extensive measurements and ray tracing across varied scenarios.It frames the compilation as covering emerging 5G system concepts and models developed worldwide.
  • Conclusion: Proper propagation models are prerequisites for advancing mmWave systems and for engineering education, with some field trials achieving 20 Gbps data rates.The conclusion connects path-loss and shadowing information with continued progress toward 5G at mmWave frequencies.
Loading 1708.02557v1…